AI Paper Atlas

Research map / Video / Video-language understanding

AVATAR: Reinforcement Learning to See, Hear, and Reason Over Video

Yogesh Kulkarni, Pooyan Fazli

CVPR 2026 · Video › Video-language understanding · approach: Event and anomaly reasoning in video · 13 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add TAS (U-shaped advantage weighting) vs uniform weights (ablation)AVATAR on Qwen2.5-Omni · OmniBenchaccuracy: 47.8 -> 49.1Improved
add TAS vs uniform weights (ablation)AVATAR on Qwen2.5-Omni · TOMATOaccuracy: 29.3 -> 30.8Improved
use TAS instead of inverse parabolic weights (ablation)AVATAR on Qwen2.5-Omni · OmniBenchaccuracy: 46.5 -> 49.1Improved
use TAS instead of linear decay weights (ablation)AVATAR on Qwen2.5-Omni · Video-Holmesaccuracy: 44.3 -> 45.1Improved
use TAS instead of linear incline weights (ablation)AVATAR on Qwen2.5-Omni · MMVUaccuracy: 65.0 -> 65.6Improved
add replay buffer to GRPO (TAS-free) (ablation)GRPO on Qwen2.5-Omni · OmniBenchaccuracy: 45.4 -> 47.8Improved
add TAS to on-policy GRPO (ablation)GRPO on Qwen2.5-Omni · VSI-Benchaccuracy: 26.5 vs 26.1 for bufferImproved
add both replay buffer and TAS to GRPO (ablation)GRPO on Qwen2.5-Omni · Video-MMMUaccuracy: 48.1 -> 49.4Improved
add off-policy replay buffer (ablation)AVATAR on Qwen2.5-Omni · WorldSense Hallucinationaccuracy: 40.1 -> 51.2Improved
add TAS (ablation)AVATAR on Qwen2.5-Omni · WorldSense Audio Changeaccuracy: 35.8 -> 37.3Improved
add SFT cold start (ablation)AVATAR on Qwen2.5-Omni · OmniBenchaccuracy: 43.8 -> 49.1Improved
add SFT cold start (ablation)GRPO on Qwen2.5-Omni · OmniBenchaccuracy: 41.2 -> 46.0Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy